Optimized Recovery Algorithms for RDP $(p, 3)$ Code

This paper is concerned with the extended RDP codes, RDP(p, 3), which can tolerate up to three concurrent disk failures. For single-disk failure recovery, we derive a lower bound on disk reads. To approach the derived lower bound, we propose an algorithm that aims at maximizing the overlapping symbols among different types of parity set. For double-disk failure recovery, we propose an algorithm to accelerate the recovery process by shortening the recovery chains with the help of the extra parities. Numerical results for p ≤ 47 show that the proposed algorithm can further reduce by approximately 10% disk reads for single-disk recovery.